In this first iteration we provide a variety of tools to both understand the metadata of your Exasol database as well as issue read queries. There is lots to explore and test and we want YOUR feedback. Please try it out and let us know what you find! We will continue to roll out more functionality in the coming months to advance more in the AI & agentic space. Your feedback will help us prioritize and identify these opportunities!
We’ve played around with this a bit. It looks good!
Our feedback so far:
The enforcement of read-only statements is good for starters. In the long run however it would be useful to optionally expand it to execution of scripts. This gives AI agents gated access to arbitrary write behavior.
An interaction like “@mcp:exasol:DESCRIBE schm.tbl” should ideally default to unquoted (case insensitive) lookup of schm.tbl. Currently it appears to handle them case sensitively. Although it (windsurf) was able to figure that out by itself after a bit it took some unnecessary back and forth.
Our biggest question mark currently is auth. The existing user/pwd approach won’t really work for us. The official MCP draft spec calls for optional oauth support for MCP servers so perhaps if you could allow the Exasol MCP server to passthrough oauth tokens to Exasol for authentication that would be great (we use oauth/OpenID for Exasol auth)
Thank you for the fantastic feedback Peter! Some thoughts:
Yes - we plan to expand the execution functionality over time. We’re trying to balance/manage the risk this inherently causes - but for true ‘agent’ functionality we definitely need to add write functionality (in a controlled manner)
Good call - I’ve added it to our list of planned improvements to improve case management/sensitivity
Oauth is next on our list actually! We’ll let you know once we’ve shipped the update - will definitely want your feedback. I’m not sure if pass through will be in the first iteration or not - we’re still scoping